date stringdate 2026-10-22 00:00:00 2026-10-23 00:00:00 | model stringclasses 1
value | hf_repo stringclasses 1
value | total_params stringclasses 1
value | active_params stringclasses 1
value | quant stringclasses 1
value | runtime stringclasses 1
value | hardware stringclasses 1
value | memory_gb int64 64 64 | test stringclasses 4
values | metric stringclasses 7
values | value stringclasses 7
values | unit stringclasses 6
values | source_post stringclasses 2
values |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
2026-10-22 | Flint | RecursiveSI/flint | 40B | 2.5B | 4-bit | Ollama | M5 Pro Mac mini | 64 | short coding prompt | decode_speed | 115 | tok/s | https://x.com/pratikg/status/2113047879548105131 |
2026-10-23 | Flint | RecursiveSI/flint | 40B | 2.5B | 4-bit | Ollama | M5 Pro Mac mini | 64 | read 20K-token codebase then add a feature | prefill_time | 11 | s | https://x.com/pratikg/status/2113087627503256038 |
2026-10-23 | Flint | RecursiveSI/flint | 40B | 2.5B | 4-bit | Ollama | M5 Pro Mac mini | 64 | read 20K-token codebase then add a feature | decode_speed_at_20k_context | 87 | tok/s | https://x.com/pratikg/status/2113087627503256038 |
2026-10-23 | Flint | RecursiveSI/flint | 40B | 2.5B | 4-bit | Ollama | M5 Pro Mac mini | 64 | read 20K-token codebase then add a feature | tests_passed | 40/40 | tests | https://x.com/pratikg/status/2113087627503256038 |
2026-10-23 | Flint | RecursiveSI/flint | 40B | 2.5B | 4-bit | Ollama | M5 Pro Mac mini | 64 | JSON parser from scratch | first_draft_test_failures | 13 | failures | https://x.com/pratikg/status/2113087627503256038 |
2026-10-23 | Flint | RecursiveSI/flint | 40B | 2.5B | 4-bit | Ollama | M5 Pro Mac mini | 64 | JSON parser from scratch | rounds_to_fix_all | 1 | rounds | https://x.com/pratikg/status/2113087627503256038 |
2026-10-23 | Flint | RecursiveSI/flint | 40B | 2.5B | 4-bit | Ollama | M5 Pro Mac mini | 64 | long coding runs | max_reasoning_tokens_before_code | 20000+ | tokens | https://x.com/pratikg/status/2113087627503256038 |
Mac mini local AI benchmarks
Real numbers from open-weight models running fully local on one machine: an M5 Pro Mac mini with 64GB of unified memory.
Every row is one measurement: model, quant, runtime, the test, and the result, with a link to the post where it was shared. New models get added as I test them.
What's in it so far
| Model | Quant | Runtime | Test | Result |
|---|---|---|---|---|
| Flint 40B-A2.5B | 4-bit | Ollama | Short coding prompt | 115 tok/s |
| Flint | 4-bit | Ollama | Read a 20K-token codebase | 11 s prefill, then 87 tok/s |
| Flint | 4-bit | Ollama | Add a feature to that codebase | 40/40 tests passed |
| Flint | 4-bit | Ollama | JSON parser from scratch | 13 failures in the first draft, all fixed in one round |
How it's measured
flint_bench.py times any OpenAI-compatible local server (Ollama, etc.) with only the Python standard library. It reports time to first token, reasoning share, and decode tok/s, as the median of 3 runs.
Coming next
The 11 models from my weekend run on the same Mac.
By @pratikg.
- Downloads last month
- 19